Translators Raging Against Machines, Part 2
I’ve seen lots of bad translations before but this time the errors were quite consistently eyebrow-raising. I actually couldn’t imagine why anyone would write what I saw. It turned out the text was a poorly edited machine translation.
What’s going on in the translation business?
Deep learning and AI brought a great deal of developments to translation services. Sadly, it seems no one has ever considered that NOT ONLY machines are to learn. Translators (or post-editors) are not prefabricated, ready-to-use androids produced by universities. They keep learning for the whole of their careers. More specifically, above others, they learn from what they read and translate.
Is active contact with texts in foreign languages still a good learning source?
Translators gain knowledge all the time. While reading, watching TV, chatting with others, and while translating. Whatever is new to the translator, has to be researched in dictionaries, expert literature, other sources. This of course requires some effort, so new terms naturally find their spots in the translator’s memory and/or private glossary. However, during the post-editing process everything is already “translated.” And that’s a problem.
I’ve already seen this – therefore I know it.
In the course of machine translation post-processing (MTPE), errors may go undetected when the post-editor is not acquainted with the terminology enough to notice that the machine got it wrong. Moreover, the machine may repeatedly expose the post-editor to the same wrong translation, initiating the human learning process. When you encounter something 5 or more times, you feel acquainted with that, like with an old friend. Your memory tricks you – something looks familiar, so it “must be OK.” Instead of a corrected text, we acquire a human who has just learned a wrong, machine-made translation.
Machines have learned from us already, what about us – humans?
While machines already have the data and keep on learning, every translator once starts from scratch. While my generation of translators developed knowledge on man-made texts and dictionaries compiled meticulously by professional linguists, the generation of post-editors will source their knowledge from machines, too. Including the errors. There is a fair risk that their resulting outputs will be recycled by machines, then by humans, and so on.
An endless, unbreakable cycle of error.